基于机器视觉的淡水鱼外形检测方法  

Freshwater fish shape detection method based on machine vision

作　　者：马商牟司 胡志刚 曾山 李皞 MA Shangmusi;HU Zhigang;ZENG Shan;LI Hao(College of Mechanical Engineering,Wuhan Polytechnic University, Wuhan Hubei 430023,China;College of Mathematics and Computer Science,Wuhan Polytechnic University,Wuhan Hubei 430023,China)

机构地区：[1]武汉轻工大学机械工程学院,武汉430023 [2]武汉轻工大学数学与计算机学院,武汉430023

出　　处：《计算机应用》2021年第S02期270-274,共5页journal of Computer Applications

基　　金：湖北省技术创新专项重大项目(2019AB085);湖北省高等学校优秀中青年科技创新团队计划项目(T2021009);湖北省自然科学杰出青年基金资助项目(2020CFA063)。

摘　　要：为了满足大宗淡水鱼智能加工生产的需要,实现对不锈钢链板上淡水鱼的轮廓检测,提出一种抗畸变性强的轮廓检测方法。该方法由畸变矫正、背景差分、椭圆拟合三部分组成。首先,采集链板上的淡水鱼图像与模板图像,构造待检测图像与标准模板图像对应像素点的投影矩阵,将畸变的待检测图像矫正至统一的坐标系下;然后,与标准模板图像匹配,对感兴趣区域(ROI)差分去除背景,得到目标图像并提取其轮廓点;最后,采用逐步椭圆拟合法,确保鱼身主体部分被精确地拟合,且长轴方向与鱼体方向一致。通过将匹配算法与背景差分法相结合,确保了校准后的待检测图像与标准模板图像的ROI完全重合;再利用逐步舍弃部分轮廓样本点的椭圆拟合法,能较好地避免噪声点的影响,精准地描述淡水鱼轮廓。实验结果表明,相较于传统的背景差分和椭圆拟合算法,所提方法能将误差范围控制在5%以内,大幅提高了检测精度。To meet the demand of intelligent processing and production on bulk freshwater fish processing line,realize the contour detection of freshwater fish on the stainless steel chain plate,a freshwater fish contour detection algorithm to resist distortion for stainless steel chain plate image was proposed,which consists of three steps:distortion correction,background difference,and ellipse fitting. First,the freshwater fish image and the template image on the chain board were collected,the distorted test image was corrected to the world coordinate system by constructing the projection matrix of the corresponding pixel points between the test image and the template image. Second,the Region Of Interest(ROI)of the test image was matched with the template image by difference for removing the backgroud to obtain the target image,from which contour points were extracted. Finally,the stepwise ellipse fitting method was adopted to ensure that the main part of the fish body was accurately fitted,and the direction of the long axis was consistent with the direction of the fish body. By combining the matching algorithm with the background difference algorithm,the calibrated image to be detected and the ROI of the standard template image are ensured to completely conincide,the influence of noise points can be better avoided by using the ellipse fitting method that gradually discards part of the contour sample points,and the outline of freshwater fish can be accurately described. The experimental results show that in comparison to the traditional background difference and ellipse fitting algorithms,the proposed algorithm can ensure the error variance within a range of 5%,greatly improving the detection accuracy.

关 键 词：畸变矫正 背景差分 椭圆拟合 机器视觉 淡水鱼加工 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]